Projection accuracy

"Projected" is what the capacity model would have forecast from each project plan, not what was typed into Forecast (we don't have Forecast's history).

• Budget vs actual: the person's role hours in Planning/Closing (Deroy = Design, Julia = Account Management, Ariel = PM, Kade = Strategy) against the actual hours they worked on the project. This is the fairest test for finished projects.
• Where actuals come from: time tracking for anyone on it (Deroy since Oct 9, 2026; Julia's ActivityWatch rows merged with her Harvest history), and Harvest otherwise and before that. Ariel and Kade log in Harvest directly. Both refresh every night. Deroy's 2025–2026 hours were reconstructed from Drive activity, Figma history, calendar, Slack, Gmail and Adobe file dates, written into Rize, then posted to Harvest on 2026-09-19 as weekly entries tagged Define / Envision / Create / Build — the phase taken from where each day falls in that project's own timeline. They are estimates: trust the monthly and project totals, not a single week.
• Projected vs actual, Jan 1 → today: budget hours spread across canvas phase dates, compared week by week with Harvest. This tests both the amount and the timing.
• Schedule elapsed / Expected to date / Pace: for a project still running, the full budget isn't the right yardstick. The model spreads each role's budget across that project's phase dates, so "expected to date" is what the plan says should be spent by today, and pace is logged ÷ expected. 100% is on plan; 150% means burning the role's hours half again as fast as the schedule assumes. A finished project shows no pace — read "vs budget" instead.
• Timing error: of the hours actually logged, the share that landed in a different week than projected (sum of weekly gaps ÷ 2 ÷ actual). 0% = perfect weeks, and it can go above 100% when the plan put the hours in weeks where nothing was logged.